Abstract:

A central assumption in psychological research is that behavioral responses provide a direct and relatively stable readout of underlying mental processes. This assumption has supported the design of many stimulus-response paradigms, especially in trial-based perceptual and cognitive tasks. However, increasing evidence from dynamic visuomotor tasks suggests that the relationship between sensory input and behavioral output is not fixed. The same physical stimulus may give rise to different response patterns depending on task structure, response format, feedback availability, and temporal continuity. The present proposal addresses this fundamental issue by reconceptualizing the stimulus-response relationship as a dynamic mapping process rather than a static correspondence between a stimulus and an isolated response.
The major innovation of this study is to introduce a Kalman filter model, grounded in a Bayesian framework, as a quantitative tool for decomposing dynamic visuomotor behavior into separable psychological components. In this framework, the response at a given moment is not treated as an independent reflection of the current stimulus. Instead, it is modeled as the result of recursive integration between current sensory input and the previous behavioral state. The Kalman gain provides a formal estimate of how strongly the system weights current visual information relative to prior response-based prediction. Thus, the model makes it possible to distinguish perceptual encoding noise from information-integration strategy, two components that are usually confounded in raw behavioral trajectories.
This proposal advances existing work in three related ways. First, it extends continuous psychophysics by using dynamic tracking behavior to infer stable perceptual parameters. Rather than relying only on traditional trial-based discrimination thresholds, the study asks whether continuous manual and eye-movement tracking can recover comparable indices of visual sensitivity. Preliminary data suggest that parameters extracted from manual tracking are meaningfully associated with discrimination noise, whereas eye-tracking parameters show weaker correspondence, possibly due to peripheral processing and saccadic interruptions. This comparison provides a direct test of whether dynamic visuomotor tasks can serve as efficient, ecologically valid tools for perceptual measurement.
Second, the study moves beyond descriptive modeling by experimentally manipulating the information sources that enter the integration process. Perceptual learning, feedback structure, feedback uncertainty, and trajectory predictability will be systematically varied to determine how they alter model-derived parameters. For example, if perceptual learning improves visual encoding, perceptual noise should decrease and the weight assigned to current sensory input should increase. Conversely, when the stimulus trajectory becomes more predictable, observers may rely more strongly on internally generated predictions or feedback-based priors, leading to a reduction in the Kalman gain. By testing these predictions, the study aims to clarify how perceptual input, feedback information, and prior expectation jointly determine dynamic visuomotor control.
Third, the proposal links the behavioral model to neural measures. Steady-state visual evoked potentials will be used as neural markers of visual cortical processing during tracking and passive viewing. This design allows the study to ask whether perceptual parameters inferred from behavior correspond to measurable activity in visual cortex. In addition, changes in alpha-band oscillations will be examined to assess whether variations in information integration are accompanied by changes in cognitive resource allocation. This neural component is important because many existing Bayesian or Kalman-filter models of visuomotor behavior remain primarily behavioral; the present proposal explicitly tests whether model parameters have identifiable electrophysiological correlates.
The theoretical contribution of the study lies in offering a unified account of why static trial-based tasks and dynamic continuous tasks may produce apparently different behavioral signatures. Rather than assuming a strict dissociation between perceptual judgment and action control, the proposed framework treats these differences as consequences of changing information weights under different task structures. In trial-based tasks, the previous response usually contains limited useful information about the current stimulus, so behavior is dominated by current sensory evidence. In continuous tracking tasks, however, the previous response rapidly accumulates stimulus-related information and becomes a reliable predictor of the current state. The apparent difference between static and dynamic tasks can therefore be explained by a shift in integration weights rather than by assuming entirely separate processing routes.
Methodologically, this study may contribute to the development of faster and more adaptive perceptual assessment tools. Continuous tracking combined with model-based parameter estimation may provide efficient estimates of color sensitivity, stereoscopic sensitivity, and other perceptual capacities. Preliminary findings from color-vision assessment indicate that short segments of tracking data can classify color-vision deficits with high accuracy, suggesting practical value for rapid screening and individualized perceptual evaluation.
Overall, this proposal aims to build a model-based framework for understanding dynamic visuomotor control. By combining continuous behavioral measurement, Bayesian modeling, experimental manipulation of information sources, and electrophysiological validation, the study seeks to clarify how sensory input, feedback, and prior prediction are integrated over time. Its core contribution is not simply to apply the Kalman filter to psychological data, but to use it as a theoretical and methodological bridge between stimulus input, internal processing, and behavioral response.

Key words: motor control, motion-related perception, Bayesian modeling, visuomotor integration, Kalman filter